A secondary framework for small targets segmentation In Remote Sensing Images

Abstract

The automatic interpreting of small object using computer in Remote Sensing Image(RSI) is sharply limited by low resolution and the uncertainty of imaging season, leading to the results of low recognition rate and poor generalization ability. In this paper, the Erlongshan Reservoir region of Heilongjiang province is selected as research area, and a secondary segmentation framework is proposed for small objects recognition based on salience detection and Hough Transform. Firstly, the salience of particular small objects is calculated to find candidates of small objects. Next, the Hough Transform is performed on an enhanced RSI constrained by the size of small size to identify small objects from others, such as highway fragment, river fragment, house and farmland and so on. The experiments results regarding small reservoir segmentation show that the method has high robustness and generalization ability, and the idea of classification can be used to the automatic interpreting process of other kind of small objects of RSI.